d3fbb1f1caf58de61791046f74e486566fcd10a4
[linux-2.6.git] / net / ipv6 / netfilter / nf_conntrack_reasm.c
1 /*
2  * IPv6 fragment reassembly for connection tracking
3  *
4  * Copyright (C)2004 USAGI/WIDE Project
5  *
6  * Author:
7  *      Yasuyuki Kozakai @USAGI <yasuyuki.kozakai@toshiba.co.jp>
8  *
9  * Based on: net/ipv6/reassembly.c
10  *
11  * This program is free software; you can redistribute it and/or
12  * modify it under the terms of the GNU General Public License
13  * as published by the Free Software Foundation; either version
14  * 2 of the License, or (at your option) any later version.
15  */
16
17 #include <linux/errno.h>
18 #include <linux/types.h>
19 #include <linux/string.h>
20 #include <linux/socket.h>
21 #include <linux/sockios.h>
22 #include <linux/jiffies.h>
23 #include <linux/net.h>
24 #include <linux/list.h>
25 #include <linux/netdevice.h>
26 #include <linux/in6.h>
27 #include <linux/ipv6.h>
28 #include <linux/icmpv6.h>
29 #include <linux/random.h>
30 #include <linux/jhash.h>
31
32 #include <net/sock.h>
33 #include <net/snmp.h>
34
35 #include <net/ipv6.h>
36 #include <net/protocol.h>
37 #include <net/transp_v6.h>
38 #include <net/rawv6.h>
39 #include <net/ndisc.h>
40 #include <net/addrconf.h>
41 #include <linux/sysctl.h>
42 #include <linux/netfilter.h>
43 #include <linux/netfilter_ipv6.h>
44 #include <linux/kernel.h>
45 #include <linux/module.h>
46
47 #if 0
48 #define DEBUGP printk
49 #else
50 #define DEBUGP(format, args...)
51 #endif
52
53 #define NF_CT_FRAG6_HIGH_THRESH 262144 /* == 256*1024 */
54 #define NF_CT_FRAG6_LOW_THRESH 196608  /* == 192*1024 */
55 #define NF_CT_FRAG6_TIMEOUT IPV6_FRAG_TIMEOUT
56
57 unsigned int nf_ct_frag6_high_thresh __read_mostly = 256*1024;
58 unsigned int nf_ct_frag6_low_thresh __read_mostly = 192*1024;
59 unsigned long nf_ct_frag6_timeout __read_mostly = IPV6_FRAG_TIMEOUT;
60
61 struct nf_ct_frag6_skb_cb
62 {
63         struct inet6_skb_parm   h;
64         int                     offset;
65         struct sk_buff          *orig;
66 };
67
68 #define NFCT_FRAG6_CB(skb)      ((struct nf_ct_frag6_skb_cb*)((skb)->cb))
69
70 struct nf_ct_frag6_queue
71 {
72         struct hlist_node       list;
73         struct list_head        lru_list;       /* lru list member      */
74
75         __be32                  id;             /* fragment id          */
76         struct in6_addr         saddr;
77         struct in6_addr         daddr;
78
79         spinlock_t              lock;
80         atomic_t                refcnt;
81         struct timer_list       timer;          /* expire timer         */
82         struct sk_buff          *fragments;
83         int                     len;
84         int                     meat;
85         ktime_t                 stamp;
86         unsigned int            csum;
87         __u8                    last_in;        /* has first/last segment arrived? */
88 #define COMPLETE                4
89 #define FIRST_IN                2
90 #define LAST_IN                 1
91         __u16                   nhoffset;
92 };
93
94 /* Hash table. */
95
96 #define FRAG6Q_HASHSZ   64
97
98 static struct hlist_head nf_ct_frag6_hash[FRAG6Q_HASHSZ];
99 static DEFINE_RWLOCK(nf_ct_frag6_lock);
100 static u32 nf_ct_frag6_hash_rnd;
101 static LIST_HEAD(nf_ct_frag6_lru_list);
102 int nf_ct_frag6_nqueues = 0;
103
104 static __inline__ void __fq_unlink(struct nf_ct_frag6_queue *fq)
105 {
106         hlist_del(&fq->list);
107         list_del(&fq->lru_list);
108         nf_ct_frag6_nqueues--;
109 }
110
111 static __inline__ void fq_unlink(struct nf_ct_frag6_queue *fq)
112 {
113         write_lock(&nf_ct_frag6_lock);
114         __fq_unlink(fq);
115         write_unlock(&nf_ct_frag6_lock);
116 }
117
118 static unsigned int ip6qhashfn(__be32 id, struct in6_addr *saddr,
119                                struct in6_addr *daddr)
120 {
121         u32 a, b, c;
122
123         a = (__force u32)saddr->s6_addr32[0];
124         b = (__force u32)saddr->s6_addr32[1];
125         c = (__force u32)saddr->s6_addr32[2];
126
127         a += JHASH_GOLDEN_RATIO;
128         b += JHASH_GOLDEN_RATIO;
129         c += nf_ct_frag6_hash_rnd;
130         __jhash_mix(a, b, c);
131
132         a += (__force u32)saddr->s6_addr32[3];
133         b += (__force u32)daddr->s6_addr32[0];
134         c += (__force u32)daddr->s6_addr32[1];
135         __jhash_mix(a, b, c);
136
137         a += (__force u32)daddr->s6_addr32[2];
138         b += (__force u32)daddr->s6_addr32[3];
139         c += (__force u32)id;
140         __jhash_mix(a, b, c);
141
142         return c & (FRAG6Q_HASHSZ - 1);
143 }
144
145 static struct timer_list nf_ct_frag6_secret_timer;
146 int nf_ct_frag6_secret_interval = 10 * 60 * HZ;
147
148 static void nf_ct_frag6_secret_rebuild(unsigned long dummy)
149 {
150         unsigned long now = jiffies;
151         int i;
152
153         write_lock(&nf_ct_frag6_lock);
154         get_random_bytes(&nf_ct_frag6_hash_rnd, sizeof(u32));
155         for (i = 0; i < FRAG6Q_HASHSZ; i++) {
156                 struct nf_ct_frag6_queue *q;
157                 struct hlist_node *p, *n;
158
159                 hlist_for_each_entry_safe(q, p, n, &nf_ct_frag6_hash[i], list) {
160                         unsigned int hval = ip6qhashfn(q->id,
161                                                        &q->saddr,
162                                                        &q->daddr);
163                         if (hval != i) {
164                                 hlist_del(&q->list);
165                                 /* Relink to new hash chain. */
166                                 hlist_add_head(&q->list,
167                                                &nf_ct_frag6_hash[hval]);
168                         }
169                 }
170         }
171         write_unlock(&nf_ct_frag6_lock);
172
173         mod_timer(&nf_ct_frag6_secret_timer, now + nf_ct_frag6_secret_interval);
174 }
175
176 atomic_t nf_ct_frag6_mem = ATOMIC_INIT(0);
177
178 /* Memory Tracking Functions. */
179 static inline void frag_kfree_skb(struct sk_buff *skb, unsigned int *work)
180 {
181         if (work)
182                 *work -= skb->truesize;
183         atomic_sub(skb->truesize, &nf_ct_frag6_mem);
184         if (NFCT_FRAG6_CB(skb)->orig)
185                 kfree_skb(NFCT_FRAG6_CB(skb)->orig);
186
187         kfree_skb(skb);
188 }
189
190 static inline void frag_free_queue(struct nf_ct_frag6_queue *fq,
191                                    unsigned int *work)
192 {
193         if (work)
194                 *work -= sizeof(struct nf_ct_frag6_queue);
195         atomic_sub(sizeof(struct nf_ct_frag6_queue), &nf_ct_frag6_mem);
196         kfree(fq);
197 }
198
199 static inline struct nf_ct_frag6_queue *frag_alloc_queue(void)
200 {
201         struct nf_ct_frag6_queue *fq = kmalloc(sizeof(struct nf_ct_frag6_queue), GFP_ATOMIC);
202
203         if (!fq)
204                 return NULL;
205         atomic_add(sizeof(struct nf_ct_frag6_queue), &nf_ct_frag6_mem);
206         return fq;
207 }
208
209 /* Destruction primitives. */
210
211 /* Complete destruction of fq. */
212 static void nf_ct_frag6_destroy(struct nf_ct_frag6_queue *fq,
213                                 unsigned int *work)
214 {
215         struct sk_buff *fp;
216
217         BUG_TRAP(fq->last_in&COMPLETE);
218         BUG_TRAP(del_timer(&fq->timer) == 0);
219
220         /* Release all fragment data. */
221         fp = fq->fragments;
222         while (fp) {
223                 struct sk_buff *xp = fp->next;
224
225                 frag_kfree_skb(fp, work);
226                 fp = xp;
227         }
228
229         frag_free_queue(fq, work);
230 }
231
232 static __inline__ void fq_put(struct nf_ct_frag6_queue *fq, unsigned int *work)
233 {
234         if (atomic_dec_and_test(&fq->refcnt))
235                 nf_ct_frag6_destroy(fq, work);
236 }
237
238 /* Kill fq entry. It is not destroyed immediately,
239  * because caller (and someone more) holds reference count.
240  */
241 static __inline__ void fq_kill(struct nf_ct_frag6_queue *fq)
242 {
243         if (del_timer(&fq->timer))
244                 atomic_dec(&fq->refcnt);
245
246         if (!(fq->last_in & COMPLETE)) {
247                 fq_unlink(fq);
248                 atomic_dec(&fq->refcnt);
249                 fq->last_in |= COMPLETE;
250         }
251 }
252
253 static void nf_ct_frag6_evictor(void)
254 {
255         struct nf_ct_frag6_queue *fq;
256         struct list_head *tmp;
257         unsigned int work;
258
259         work = atomic_read(&nf_ct_frag6_mem);
260         if (work <= nf_ct_frag6_low_thresh)
261                 return;
262
263         work -= nf_ct_frag6_low_thresh;
264         while (work > 0) {
265                 read_lock(&nf_ct_frag6_lock);
266                 if (list_empty(&nf_ct_frag6_lru_list)) {
267                         read_unlock(&nf_ct_frag6_lock);
268                         return;
269                 }
270                 tmp = nf_ct_frag6_lru_list.next;
271                 BUG_ON(tmp == NULL);
272                 fq = list_entry(tmp, struct nf_ct_frag6_queue, lru_list);
273                 atomic_inc(&fq->refcnt);
274                 read_unlock(&nf_ct_frag6_lock);
275
276                 spin_lock(&fq->lock);
277                 if (!(fq->last_in&COMPLETE))
278                         fq_kill(fq);
279                 spin_unlock(&fq->lock);
280
281                 fq_put(fq, &work);
282         }
283 }
284
285 static void nf_ct_frag6_expire(unsigned long data)
286 {
287         struct nf_ct_frag6_queue *fq = (struct nf_ct_frag6_queue *) data;
288
289         spin_lock(&fq->lock);
290
291         if (fq->last_in & COMPLETE)
292                 goto out;
293
294         fq_kill(fq);
295
296 out:
297         spin_unlock(&fq->lock);
298         fq_put(fq, NULL);
299 }
300
301 /* Creation primitives. */
302
303 static struct nf_ct_frag6_queue *nf_ct_frag6_intern(unsigned int hash,
304                                           struct nf_ct_frag6_queue *fq_in)
305 {
306         struct nf_ct_frag6_queue *fq;
307 #ifdef CONFIG_SMP
308         struct hlist_node *n;
309 #endif
310
311         write_lock(&nf_ct_frag6_lock);
312 #ifdef CONFIG_SMP
313         hlist_for_each_entry(fq, n, &nf_ct_frag6_hash[hash], list) {
314                 if (fq->id == fq_in->id &&
315                     ipv6_addr_equal(&fq_in->saddr, &fq->saddr) &&
316                     ipv6_addr_equal(&fq_in->daddr, &fq->daddr)) {
317                         atomic_inc(&fq->refcnt);
318                         write_unlock(&nf_ct_frag6_lock);
319                         fq_in->last_in |= COMPLETE;
320                         fq_put(fq_in, NULL);
321                         return fq;
322                 }
323         }
324 #endif
325         fq = fq_in;
326
327         if (!mod_timer(&fq->timer, jiffies + nf_ct_frag6_timeout))
328                 atomic_inc(&fq->refcnt);
329
330         atomic_inc(&fq->refcnt);
331         hlist_add_head(&fq->list, &nf_ct_frag6_hash[hash]);
332         INIT_LIST_HEAD(&fq->lru_list);
333         list_add_tail(&fq->lru_list, &nf_ct_frag6_lru_list);
334         nf_ct_frag6_nqueues++;
335         write_unlock(&nf_ct_frag6_lock);
336         return fq;
337 }
338
339
340 static struct nf_ct_frag6_queue *
341 nf_ct_frag6_create(unsigned int hash, __be32 id, struct in6_addr *src,                             struct in6_addr *dst)
342 {
343         struct nf_ct_frag6_queue *fq;
344
345         if ((fq = frag_alloc_queue()) == NULL) {
346                 DEBUGP("Can't alloc new queue\n");
347                 goto oom;
348         }
349
350         memset(fq, 0, sizeof(struct nf_ct_frag6_queue));
351
352         fq->id = id;
353         ipv6_addr_copy(&fq->saddr, src);
354         ipv6_addr_copy(&fq->daddr, dst);
355
356         init_timer(&fq->timer);
357         fq->timer.function = nf_ct_frag6_expire;
358         fq->timer.data = (long) fq;
359         spin_lock_init(&fq->lock);
360         atomic_set(&fq->refcnt, 1);
361
362         return nf_ct_frag6_intern(hash, fq);
363
364 oom:
365         return NULL;
366 }
367
368 static __inline__ struct nf_ct_frag6_queue *
369 fq_find(__be32 id, struct in6_addr *src, struct in6_addr *dst)
370 {
371         struct nf_ct_frag6_queue *fq;
372         struct hlist_node *n;
373         unsigned int hash = ip6qhashfn(id, src, dst);
374
375         read_lock(&nf_ct_frag6_lock);
376         hlist_for_each_entry(fq, n, &nf_ct_frag6_hash[hash], list) {
377                 if (fq->id == id &&
378                     ipv6_addr_equal(src, &fq->saddr) &&
379                     ipv6_addr_equal(dst, &fq->daddr)) {
380                         atomic_inc(&fq->refcnt);
381                         read_unlock(&nf_ct_frag6_lock);
382                         return fq;
383                 }
384         }
385         read_unlock(&nf_ct_frag6_lock);
386
387         return nf_ct_frag6_create(hash, id, src, dst);
388 }
389
390
391 static int nf_ct_frag6_queue(struct nf_ct_frag6_queue *fq, struct sk_buff *skb,
392                              struct frag_hdr *fhdr, int nhoff)
393 {
394         struct sk_buff *prev, *next;
395         int offset, end;
396
397         if (fq->last_in & COMPLETE) {
398                 DEBUGP("Allready completed\n");
399                 goto err;
400         }
401
402         offset = ntohs(fhdr->frag_off) & ~0x7;
403         end = offset + (ntohs(ipv6_hdr(skb)->payload_len) -
404                         ((u8 *)(fhdr + 1) - (u8 *)(ipv6_hdr(skb) + 1)));
405
406         if ((unsigned int)end > IPV6_MAXPLEN) {
407                 DEBUGP("offset is too large.\n");
408                 return -1;
409         }
410
411         if (skb->ip_summed == CHECKSUM_COMPLETE) {
412                 const unsigned char *nh = skb_network_header(skb);
413                 skb->csum = csum_sub(skb->csum,
414                                      csum_partial(nh, (u8 *)(fhdr + 1) - nh,
415                                                   0));
416         }
417
418         /* Is this the final fragment? */
419         if (!(fhdr->frag_off & htons(IP6_MF))) {
420                 /* If we already have some bits beyond end
421                  * or have different end, the segment is corrupted.
422                  */
423                 if (end < fq->len ||
424                     ((fq->last_in & LAST_IN) && end != fq->len)) {
425                         DEBUGP("already received last fragment\n");
426                         goto err;
427                 }
428                 fq->last_in |= LAST_IN;
429                 fq->len = end;
430         } else {
431                 /* Check if the fragment is rounded to 8 bytes.
432                  * Required by the RFC.
433                  */
434                 if (end & 0x7) {
435                         /* RFC2460 says always send parameter problem in
436                          * this case. -DaveM
437                          */
438                         DEBUGP("the end of this fragment is not rounded to 8 bytes.\n");
439                         return -1;
440                 }
441                 if (end > fq->len) {
442                         /* Some bits beyond end -> corruption. */
443                         if (fq->last_in & LAST_IN) {
444                                 DEBUGP("last packet already reached.\n");
445                                 goto err;
446                         }
447                         fq->len = end;
448                 }
449         }
450
451         if (end == offset)
452                 goto err;
453
454         /* Point into the IP datagram 'data' part. */
455         if (!pskb_pull(skb, (u8 *) (fhdr + 1) - skb->data)) {
456                 DEBUGP("queue: message is too short.\n");
457                 goto err;
458         }
459         if (pskb_trim_rcsum(skb, end - offset)) {
460                 DEBUGP("Can't trim\n");
461                 goto err;
462         }
463
464         /* Find out which fragments are in front and at the back of us
465          * in the chain of fragments so far.  We must know where to put
466          * this fragment, right?
467          */
468         prev = NULL;
469         for (next = fq->fragments; next != NULL; next = next->next) {
470                 if (NFCT_FRAG6_CB(next)->offset >= offset)
471                         break;  /* bingo! */
472                 prev = next;
473         }
474
475         /* We found where to put this one.  Check for overlap with
476          * preceding fragment, and, if needed, align things so that
477          * any overlaps are eliminated.
478          */
479         if (prev) {
480                 int i = (NFCT_FRAG6_CB(prev)->offset + prev->len) - offset;
481
482                 if (i > 0) {
483                         offset += i;
484                         if (end <= offset) {
485                                 DEBUGP("overlap\n");
486                                 goto err;
487                         }
488                         if (!pskb_pull(skb, i)) {
489                                 DEBUGP("Can't pull\n");
490                                 goto err;
491                         }
492                         if (skb->ip_summed != CHECKSUM_UNNECESSARY)
493                                 skb->ip_summed = CHECKSUM_NONE;
494                 }
495         }
496
497         /* Look for overlap with succeeding segments.
498          * If we can merge fragments, do it.
499          */
500         while (next && NFCT_FRAG6_CB(next)->offset < end) {
501                 /* overlap is 'i' bytes */
502                 int i = end - NFCT_FRAG6_CB(next)->offset;
503
504                 if (i < next->len) {
505                         /* Eat head of the next overlapped fragment
506                          * and leave the loop. The next ones cannot overlap.
507                          */
508                         DEBUGP("Eat head of the overlapped parts.: %d", i);
509                         if (!pskb_pull(next, i))
510                                 goto err;
511
512                         /* next fragment */
513                         NFCT_FRAG6_CB(next)->offset += i;
514                         fq->meat -= i;
515                         if (next->ip_summed != CHECKSUM_UNNECESSARY)
516                                 next->ip_summed = CHECKSUM_NONE;
517                         break;
518                 } else {
519                         struct sk_buff *free_it = next;
520
521                         /* Old fragmnet is completely overridden with
522                          * new one drop it.
523                          */
524                         next = next->next;
525
526                         if (prev)
527                                 prev->next = next;
528                         else
529                                 fq->fragments = next;
530
531                         fq->meat -= free_it->len;
532                         frag_kfree_skb(free_it, NULL);
533                 }
534         }
535
536         NFCT_FRAG6_CB(skb)->offset = offset;
537
538         /* Insert this fragment in the chain of fragments. */
539         skb->next = next;
540         if (prev)
541                 prev->next = skb;
542         else
543                 fq->fragments = skb;
544
545         skb->dev = NULL;
546         fq->stamp = skb->tstamp;
547         fq->meat += skb->len;
548         atomic_add(skb->truesize, &nf_ct_frag6_mem);
549
550         /* The first fragment.
551          * nhoffset is obtained from the first fragment, of course.
552          */
553         if (offset == 0) {
554                 fq->nhoffset = nhoff;
555                 fq->last_in |= FIRST_IN;
556         }
557         write_lock(&nf_ct_frag6_lock);
558         list_move_tail(&fq->lru_list, &nf_ct_frag6_lru_list);
559         write_unlock(&nf_ct_frag6_lock);
560         return 0;
561
562 err:
563         return -1;
564 }
565
566 /*
567  *      Check if this packet is complete.
568  *      Returns NULL on failure by any reason, and pointer
569  *      to current nexthdr field in reassembled frame.
570  *
571  *      It is called with locked fq, and caller must check that
572  *      queue is eligible for reassembly i.e. it is not COMPLETE,
573  *      the last and the first frames arrived and all the bits are here.
574  */
575 static struct sk_buff *
576 nf_ct_frag6_reasm(struct nf_ct_frag6_queue *fq, struct net_device *dev)
577 {
578         struct sk_buff *fp, *op, *head = fq->fragments;
579         int    payload_len;
580
581         fq_kill(fq);
582
583         BUG_TRAP(head != NULL);
584         BUG_TRAP(NFCT_FRAG6_CB(head)->offset == 0);
585
586         /* Unfragmented part is taken from the first segment. */
587         payload_len = ((head->data - skb_network_header(head)) -
588                        sizeof(struct ipv6hdr) + fq->len -
589                        sizeof(struct frag_hdr));
590         if (payload_len > IPV6_MAXPLEN) {
591                 DEBUGP("payload len is too large.\n");
592                 goto out_oversize;
593         }
594
595         /* Head of list must not be cloned. */
596         if (skb_cloned(head) && pskb_expand_head(head, 0, 0, GFP_ATOMIC)) {
597                 DEBUGP("skb is cloned but can't expand head");
598                 goto out_oom;
599         }
600
601         /* If the first fragment is fragmented itself, we split
602          * it to two chunks: the first with data and paged part
603          * and the second, holding only fragments. */
604         if (skb_shinfo(head)->frag_list) {
605                 struct sk_buff *clone;
606                 int i, plen = 0;
607
608                 if ((clone = alloc_skb(0, GFP_ATOMIC)) == NULL) {
609                         DEBUGP("Can't alloc skb\n");
610                         goto out_oom;
611                 }
612                 clone->next = head->next;
613                 head->next = clone;
614                 skb_shinfo(clone)->frag_list = skb_shinfo(head)->frag_list;
615                 skb_shinfo(head)->frag_list = NULL;
616                 for (i=0; i<skb_shinfo(head)->nr_frags; i++)
617                         plen += skb_shinfo(head)->frags[i].size;
618                 clone->len = clone->data_len = head->data_len - plen;
619                 head->data_len -= clone->len;
620                 head->len -= clone->len;
621                 clone->csum = 0;
622                 clone->ip_summed = head->ip_summed;
623
624                 NFCT_FRAG6_CB(clone)->orig = NULL;
625                 atomic_add(clone->truesize, &nf_ct_frag6_mem);
626         }
627
628         /* We have to remove fragment header from datagram and to relocate
629          * header in order to calculate ICV correctly. */
630         skb_network_header(head)[fq->nhoffset] = head->h.raw[0];
631         memmove(head->head + sizeof(struct frag_hdr), head->head,
632                 (head->data - head->head) - sizeof(struct frag_hdr));
633         head->mac.raw += sizeof(struct frag_hdr);
634         head->nh.raw += sizeof(struct frag_hdr);
635
636         skb_shinfo(head)->frag_list = head->next;
637         head->h.raw = head->data;
638         skb_push(head, head->data - skb_network_header(head));
639         atomic_sub(head->truesize, &nf_ct_frag6_mem);
640
641         for (fp=head->next; fp; fp = fp->next) {
642                 head->data_len += fp->len;
643                 head->len += fp->len;
644                 if (head->ip_summed != fp->ip_summed)
645                         head->ip_summed = CHECKSUM_NONE;
646                 else if (head->ip_summed == CHECKSUM_COMPLETE)
647                         head->csum = csum_add(head->csum, fp->csum);
648                 head->truesize += fp->truesize;
649                 atomic_sub(fp->truesize, &nf_ct_frag6_mem);
650         }
651
652         head->next = NULL;
653         head->dev = dev;
654         head->tstamp = fq->stamp;
655         ipv6_hdr(head)->payload_len = htons(payload_len);
656
657         /* Yes, and fold redundant checksum back. 8) */
658         if (head->ip_summed == CHECKSUM_COMPLETE)
659                 head->csum = csum_partial(skb_network_header(head),
660                                           head->h.raw - head->nh.raw,
661                                           head->csum);
662
663         fq->fragments = NULL;
664
665         /* all original skbs are linked into the NFCT_FRAG6_CB(head).orig */
666         fp = skb_shinfo(head)->frag_list;
667         if (NFCT_FRAG6_CB(fp)->orig == NULL)
668                 /* at above code, head skb is divided into two skbs. */
669                 fp = fp->next;
670
671         op = NFCT_FRAG6_CB(head)->orig;
672         for (; fp; fp = fp->next) {
673                 struct sk_buff *orig = NFCT_FRAG6_CB(fp)->orig;
674
675                 op->next = orig;
676                 op = orig;
677                 NFCT_FRAG6_CB(fp)->orig = NULL;
678         }
679
680         return head;
681
682 out_oversize:
683         if (net_ratelimit())
684                 printk(KERN_DEBUG "nf_ct_frag6_reasm: payload len = %d\n", payload_len);
685         goto out_fail;
686 out_oom:
687         if (net_ratelimit())
688                 printk(KERN_DEBUG "nf_ct_frag6_reasm: no memory for reassembly\n");
689 out_fail:
690         return NULL;
691 }
692
693 /*
694  * find the header just before Fragment Header.
695  *
696  * if success return 0 and set ...
697  * (*prevhdrp): the value of "Next Header Field" in the header
698  *              just before Fragment Header.
699  * (*prevhoff): the offset of "Next Header Field" in the header
700  *              just before Fragment Header.
701  * (*fhoff)   : the offset of Fragment Header.
702  *
703  * Based on ipv6_skip_hdr() in net/ipv6/exthdr.c
704  *
705  */
706 static int
707 find_prev_fhdr(struct sk_buff *skb, u8 *prevhdrp, int *prevhoff, int *fhoff)
708 {
709         u8 nexthdr = ipv6_hdr(skb)->nexthdr;
710         u8 prev_nhoff = (u8 *)&ipv6_hdr(skb)->nexthdr - skb->data;
711         int start = (u8 *)(ipv6_hdr(skb) + 1) - skb->data;
712         int len = skb->len - start;
713         u8 prevhdr = NEXTHDR_IPV6;
714
715         while (nexthdr != NEXTHDR_FRAGMENT) {
716                 struct ipv6_opt_hdr hdr;
717                 int hdrlen;
718
719                 if (!ipv6_ext_hdr(nexthdr)) {
720                         return -1;
721                 }
722                 if (len < (int)sizeof(struct ipv6_opt_hdr)) {
723                         DEBUGP("too short\n");
724                         return -1;
725                 }
726                 if (nexthdr == NEXTHDR_NONE) {
727                         DEBUGP("next header is none\n");
728                         return -1;
729                 }
730                 if (skb_copy_bits(skb, start, &hdr, sizeof(hdr)))
731                         BUG();
732                 if (nexthdr == NEXTHDR_AUTH)
733                         hdrlen = (hdr.hdrlen+2)<<2;
734                 else
735                         hdrlen = ipv6_optlen(&hdr);
736
737                 prevhdr = nexthdr;
738                 prev_nhoff = start;
739
740                 nexthdr = hdr.nexthdr;
741                 len -= hdrlen;
742                 start += hdrlen;
743         }
744
745         if (len < 0)
746                 return -1;
747
748         *prevhdrp = prevhdr;
749         *prevhoff = prev_nhoff;
750         *fhoff = start;
751
752         return 0;
753 }
754
755 struct sk_buff *nf_ct_frag6_gather(struct sk_buff *skb)
756 {
757         struct sk_buff *clone;
758         struct net_device *dev = skb->dev;
759         struct frag_hdr *fhdr;
760         struct nf_ct_frag6_queue *fq;
761         struct ipv6hdr *hdr;
762         int fhoff, nhoff;
763         u8 prevhdr;
764         struct sk_buff *ret_skb = NULL;
765
766         /* Jumbo payload inhibits frag. header */
767         if (ipv6_hdr(skb)->payload_len == 0) {
768                 DEBUGP("payload len = 0\n");
769                 return skb;
770         }
771
772         if (find_prev_fhdr(skb, &prevhdr, &nhoff, &fhoff) < 0)
773                 return skb;
774
775         clone = skb_clone(skb, GFP_ATOMIC);
776         if (clone == NULL) {
777                 DEBUGP("Can't clone skb\n");
778                 return skb;
779         }
780
781         NFCT_FRAG6_CB(clone)->orig = skb;
782
783         if (!pskb_may_pull(clone, fhoff + sizeof(*fhdr))) {
784                 DEBUGP("message is too short.\n");
785                 goto ret_orig;
786         }
787
788         clone->h.raw = clone->data + fhoff;
789         hdr = ipv6_hdr(clone);
790         fhdr = (struct frag_hdr *)clone->h.raw;
791
792         if (!(fhdr->frag_off & htons(0xFFF9))) {
793                 DEBUGP("Invalid fragment offset\n");
794                 /* It is not a fragmented frame */
795                 goto ret_orig;
796         }
797
798         if (atomic_read(&nf_ct_frag6_mem) > nf_ct_frag6_high_thresh)
799                 nf_ct_frag6_evictor();
800
801         fq = fq_find(fhdr->identification, &hdr->saddr, &hdr->daddr);
802         if (fq == NULL) {
803                 DEBUGP("Can't find and can't create new queue\n");
804                 goto ret_orig;
805         }
806
807         spin_lock(&fq->lock);
808
809         if (nf_ct_frag6_queue(fq, clone, fhdr, nhoff) < 0) {
810                 spin_unlock(&fq->lock);
811                 DEBUGP("Can't insert skb to queue\n");
812                 fq_put(fq, NULL);
813                 goto ret_orig;
814         }
815
816         if (fq->last_in == (FIRST_IN|LAST_IN) && fq->meat == fq->len) {
817                 ret_skb = nf_ct_frag6_reasm(fq, dev);
818                 if (ret_skb == NULL)
819                         DEBUGP("Can't reassemble fragmented packets\n");
820         }
821         spin_unlock(&fq->lock);
822
823         fq_put(fq, NULL);
824         return ret_skb;
825
826 ret_orig:
827         kfree_skb(clone);
828         return skb;
829 }
830
831 void nf_ct_frag6_output(unsigned int hooknum, struct sk_buff *skb,
832                         struct net_device *in, struct net_device *out,
833                         int (*okfn)(struct sk_buff *))
834 {
835         struct sk_buff *s, *s2;
836
837         for (s = NFCT_FRAG6_CB(skb)->orig; s;) {
838                 nf_conntrack_put_reasm(s->nfct_reasm);
839                 nf_conntrack_get_reasm(skb);
840                 s->nfct_reasm = skb;
841
842                 s2 = s->next;
843                 s->next = NULL;
844
845                 NF_HOOK_THRESH(PF_INET6, hooknum, s, in, out, okfn,
846                                NF_IP6_PRI_CONNTRACK_DEFRAG + 1);
847                 s = s2;
848         }
849         nf_conntrack_put_reasm(skb);
850 }
851
852 int nf_ct_frag6_kfree_frags(struct sk_buff *skb)
853 {
854         struct sk_buff *s, *s2;
855
856         for (s = NFCT_FRAG6_CB(skb)->orig; s; s = s2) {
857
858                 s2 = s->next;
859                 kfree_skb(s);
860         }
861
862         kfree_skb(skb);
863
864         return 0;
865 }
866
867 int nf_ct_frag6_init(void)
868 {
869         nf_ct_frag6_hash_rnd = (u32) ((num_physpages ^ (num_physpages>>7)) ^
870                                    (jiffies ^ (jiffies >> 6)));
871
872         init_timer(&nf_ct_frag6_secret_timer);
873         nf_ct_frag6_secret_timer.function = nf_ct_frag6_secret_rebuild;
874         nf_ct_frag6_secret_timer.expires = jiffies
875                                            + nf_ct_frag6_secret_interval;
876         add_timer(&nf_ct_frag6_secret_timer);
877
878         return 0;
879 }
880
881 void nf_ct_frag6_cleanup(void)
882 {
883         del_timer(&nf_ct_frag6_secret_timer);
884         nf_ct_frag6_low_thresh = 0;
885         nf_ct_frag6_evictor();
886 }